You are off by a couple of orders of magnitude

I have run 500+ req/sec on a raspberry pi using 4 TB dataset with 2 GB of RAM, with under 100ms for the 99.99 percentile

A few hundreds req a second is basically nothing.